[2S6-OS-7a] OS-7

[2S6-OS-7a-02] Development of a Method for Pharmacovigilance of Adverse Drug Events from Electronic Health Record

〇Kiminori Shimamoto¹, Yoshimasa Kawazoe¹, Emiko Shinohara¹, Shuntaro Yada², Shoko Wakamiya², Shungo Imai³, Satoko Hori³, Eiji Aramaki² (1. Artificial Intelligence and Digital Twin in Healthcare, Graduate School of Medicine, The University of Tokyo, 2. Social Computing Laboratory, Graduate School of Science and Technology, Nara Institute of Science and Technology, 3. Division of Drug Informatics, Keio University Faculty of Pharmacy)

Keywords:Clinical text, Adverse drug events, Named entity, Propensity score, Time to event analysis

Background: Post-marketing surveillance (PMS) is necessary to monitor the safety of pharmaceutical products. However, PMS based on spontaneous reports are pointed out to underreport adverse events, so a more comprehensive and precise detection of adverse events using electronic medical records is expected. This study will develop a set of methods to detect adverse event signals from medical texts recorded in natural language and test the detectability of known adverse events. Methods: Adverse events were extracted from the medical texts using a named entity recognition tool of Japanese medical documents developed by the co-authors. The settings for adjusting the propensity score for patient backgrounds in the intervention and control groups were examined, and the detectability of adverse events for multiple anticancer agents was tested using the Cox proportional hazards model. Results: Reasonable results were obtained for all adverse events evaluated. Discussion: We confirmed that various adverse events recorded in medical texts can be accurately extracted by natural language processing and can be a powerful source of information for PMS. In addition, the ability to handle information on patient background and time until the occurrence of an event suggests that more detailed investigation of adverse events will become possible beyond the detection of signals.